The ConTech Exec

A construction site is the most dangerous version of a building, and it's the one with the least fire protection.

I sat down with James Pecz, VP of North America at Ramtech. The Nottingham, UK company has been building wireless site-safety systems since 1990 and now runs its North American business out of Charlotte. James spent 15 years at a UK general contractor before he crossed over, so he knows what an alarm on a half-built job has to survive.

The numbers stopped me cold. NFPA counts about 4,400 construction-site fires a year in the US. They're about 1% of structure fires but nearly 4% of property damage.

We talk about:

• How a mesh radio network keeps alarms talking on sites with no Wi-Fi, no cell signal and no single point of failure
• Why the UK made temporary fire detection standard practice, and what the US is borrowing
• NFPA 241, the Fire Prevention Program Manager, and the new NFPA 72 Chapter 28 James helped write
• Whether wireless detection can stand in for a human fire watch
• The medical-alert button that saved a worker having a heart attack
• Data centers, where a fire is a schedule risk nobody can afford

Safety we take for granted is safety nobody planned for.

Guest: James Pecz, VP North America, Ramtech

Real People, Real Work, Real Talk.

Chapters:

• 00:00 Welcome ·
• 00:44 Ramtech's backstory ·
• 03:42 UK code & US adoption ·
• 05:57 Mesh networks ·
• 09:39 Evacuation & medical alerts ·
• 13:37 Mining & tunnels ·
• 14:17 Fire statistics ·
• 17:01 The phone call that started it ·
• 18:47 Reactive vs. proactive safety ·
• 20:24 NFPA 241 ·
• 21:48 NFPA 72 Ch. 28 ·
• 23:25 Replacing fire watch ·
• 25:33 Small contractors ·
• 28:57 Leasing vs. purchase ·
• 32:28 Modular builds ·
• 34:41 Success stories ·
• 37:18 Safety we take for granted ·
• 40:02 AI & data centers ·
• 42:01 Data center schedule risk ·
• 45:22 Wrap-up

Every estimator is a walking database, and some day that database retires.

I sat down with Hamid Hajian, founder of Zebel, who grew up on job sites with his dad, a developer and GC. He saw a problem in precon at a big Bay Area multifamily builder and fixed it with a spreadsheet. That spreadsheet turned into a company.

He never took VC money. When COVID shut the checkbook, he cashed out his 401k and bet on himself.

We talk about:

• Why a “2020 project” is really a 2018 project (the date you bought out the trades is the date that matters)
• Why most companies use about 5% of their historical data
• Why the estimate and the budget never seem to line up
• The AI estimating tool that shut down because nobody could explain its numbers
• Why investors kept asking “where’s the AI?” and customers kept saying that’s the reason they buy
• How much time precon deserves before the cost of capital starts eating you alive
• Why only a small slice of every AI dollar goes into construction, and why you still can’t make toast without the electrician

Building the estimate is half the job. Defending it is the other half.

Real talk with real people about real work.

Data center equipment lead times are stretching construction schedules, and generators, switchgear, and transformers are where it hurts most. Mike Powers, Co-Founder and CEO of BuildVision, walks me through a data center from the grid to the chip and shows why every piece has its own supply chain and its own waiting line.

Then he tells a story I can't shake. At a data center conference in January, everybody was designing around a 2 megawatt generator. By summer, the factory had a pile of 1.5 megawatt units sitting on its line, unsold. When lead times run 30 to 60 weeks, a miss like that stays missed.

What we get into:

• Why construction buys equipment one job at a time, while farming and car makers buy ahead
• Why a brand-new type of transformer is like buying a first-year car when the building can't go down
• Why hospital backup generators are stuck in the same line
• Mike's argument that AI will be the price of staying in the game for contractors, just like BIM, and the money will land with the owner

This is the biggest build of our time, and construction is standing in the crosshairs.

Guest: Mike Powers, Co-Founder & CEO, BuildVision
Mike on LinkedIn: https://www.linkedin.com/in/mikerpowers/
BuildVision: https://www.buildvision.io/
BuildVision on LinkedIn: https://www.linkedin.com/company/trybuildvision/

Real People, Real Work, Real Talk.

Everybody's building data centers the size of a small city. What if the better answer is a lot smaller?

This week I sat down with Meltem Ballan, CEO and co-founder of Concrete Engine. She started her career as a pattern recognition engineer, worked on NATO projects, studied the brain, and spent time at General Motors. Now she's building distributed AI "factories" that are small, modular, and placed where the power and land already exist.

This one is about where AI and construction meet. I want my construction folks to see a different model than the giant mission-critical build.

What we get into:

• Why the big cloud companies have used up most of the computing power, and what that means for researchers and startups
• Distributed AI factories: data centers of 10 megawatts or less, linked together
• Empty office floors, grain elevators, and farmland as places for data centers
• Cutting build costs with containers, insulation, and older generators
• Why power price isn't the only bottleneck, and why construction cost matters just as much
• What construction workers and local communities get out of it, beyond a few jobs
• Farmers owning their own data
• Meltem's take on open-source AI tools going closed

Smaller, local, and built with the community instead of dropped on top of it.

Find Concrete Engine at concreteengine.com
Connect with Meltem on LinkedIn: https://www.linkedin.com/in/meltemballan/

Real People, Real Work, Real Talk.

Projects don't blow up in one big moment. They drift. The master schedule says one thing, the job site is doing another, and nobody finds out until somebody walks into the Monday meeting and starts shouting.

This week I sat down with Carlos Carvalho of Aphex, calling in from London. Carlos started out with Costain on Crossrail (the Elizabeth Line) and HS2. He found the same old spreadsheets running some of the biggest jobs in the world. So he and his founding team built a planning tool for the people actually building the job, not the people reporting on it.

We get into:

• Why a spreadsheet is where a plan goes to die
• Keeping the master schedule and the daily field plan from drifting apart
• Catching delays the next day instead of weeks later
• Talking into your phone instead of typing up a daily report at the end of a long shift
• Landing a $6 billion job as your third customer, and the hair it cost him
• Why data won't replace experience, and why no decision is sometimes worse than a bad one

Carlos is moving his family to Toronto in January to build out Aphex in North America. He showed his wife the city in July at 35 degrees. January is looking more like minus 25. Good luck, brother. Texas is always an option.

Real People, Real Work, Real Talk.

Everybody thinks you plug these AI chips in like a toaster, flip the switch, and off it goes.

It doesn't work that way.

The new AI chips don't pull steady power. They spike and drop many times a second. Michael Wood III of DG Matrix tells me that at big-campus scale, those swings are cracking turbines, shearing generator shafts, and frying backup power gear in days to weeks. He says that problem has gone from theory to reality.

This one's a bit of a shift for me. I usually sit down with software founders. But the AI buildout is running into a physical wall, and I want my AEC folks to see the hardware that's out there, not just the same system that gets drawn the same way on every job.

DG Matrix builds what's called a solid state transformer. A regular transformer is a big hunk of iron, copper, and steel: one voltage in, one voltage out, about the same thing Tesla and Edison were putting in the ground.

Their version uses power electronics instead, and it takes power from several sources at once (the grid, batteries, on-site generators, backup gensets) and sends it out to several loads. Michael says one unit does the work of 10 to 15 separate pieces of gear.

What we get into:

• Why the big data center builders aren't waiting on the grid. They're making their own power on site and leaving room for the utility to tie in later.
• Same hardware on every job. The differences get set up in the software, not rebuilt in the field.
• Permitting and certification on gear the permit office has never seen before. Michael also breaks a little news about their testing lab.
• Lead times. Michael says they're typically delivering in five to six months.
• Skilled labor. He says 80 to 90 percent of the system gets built in the factory, so the site work comes down to cable terminations and commissioning.
• Texas, ERCOT, Senate Bill 6, and whether a data center can be a good neighbor to the grid instead of a risk to it.
• What these projects owe the ratepayer and the town they land in.
• Small modular data centers, and the substations around the country that still have power available.

You can throw $700 billion at a problem, but somebody still has to make it, ship it, set it, connect it, commission it, and service it for the next 20 to 30 years.

Electricians coming up right now, pay attention to this one.

Find DG Matrix at dgmatrix.com.

Real People, Real Work, Real Talk.

Every project management software out there promises "real-time visibility." Hussain Arwani was managing 15 to 20 nuclear construction projects at once, and he'll tell you straight: he didn't have it.

When the boss asked for a status, the honest answer was "let me get back to you."

So he built Masaro.

Here's the idea. Superintendents and foremen already talk their updates all day, in phone calls, schedule meetings, and end-of-shift reports. The supers Hussain interviewed told him it eats hours of their day.

Masaro lets them say it once, into their phone. The update is written out as they speak, lands on a dashboard, and moves the schedule in P6 or Microsoft Project.

Anything that pushes the critical path gets flagged for the project team to approve first. The schedule doesn't get to run off on its own.

In this episode we get into:

• Why most construction software gets built for the office (hint: the field doesn't hold the budget)
• Whether the contract schedule is a plan or a piece of leverage. I've called it fiction. Hussain was more polite.
• Why everybody wants the real-time truth from the jobsite, right up until they get it
• The number I think matters most: how long it takes to get from "we have a problem" to somebody actually doing something about it
• Who owns the project data, and why in nuclear the answer is short: the owner owns all of it
• Why a pen and a notebook on the dash of my truck is still the best cybersecurity out there

The truth will set you free. It'll piss you off first.

Masaro.io is now out of stealth,

If you back construction tech that starts in the field instead of the corner office, he's worth a call.

Find Hussain: masaro.io | Hussain Arwani on LinkedIn

Real People. Real Work. Real Talk.

Thirty-five years in this business and I've never seen this career path: DARPA robotics, then years leading Amazon's robot arms that pick and pack everything you order, then… go build houses.

Felipe Polido, co-founder and head of technology at Reframe Systems, did exactly that.

The plans an architect hands you are a rough recipe. Where the outlet goes, sure. How the wire runs, where the staples land, what gets drilled first — that's left to the trades to figure out in the field, on every single house, every single time.

Reframe's answer isn't a magic wall panel that clicks together. It's software that generates the actual build instructions, then prints them right on the lumber and the sheet goods. The stud tells you what it is. The subfloor shows you where the light goes. The electrician looks up and there's one hole, already drilled, right where the wire runs.

We get into what the robots do and don't do (the dull, dirty, dangerous framing work — not the finish carpentry), how a machine deals with twisted grade-two lumber (a real carpenter still sorts the pile first, and I had opinions about that), why they build with the same materials as everybody else instead of chasing a venture-backed miracle product, and how a three-story triple-decker went from crane on Tuesday to a building on Friday.

And then the honest part. Even with a factory that keeps getting faster, Felipe says the things that could stop them cold are the same things that have stopped all of us: permitting, foundations, and inspectors who read the code a little differently in every town. Fire, apparently, spreads differently from municipality to municipality.

Felipe's team is moving out of a 17,000-square-foot pilot facility into a building roughly seven times that size, and they're going from signed lease to operational in about seventy days. Ask a factory owner how easy that is.

Real People, Real Work, Real Talk.

There's an excavator loading trucks on a live job site right now with nobody in the cab.

Not a demo. Not a parking lot test. A paying job with 1.2 million yards of dirt to move, and that's just phase one.

Rachael Ferrera leads innovation at Zachry Construction. Zachry, Sundt Construction and Champion Site Prep are the first three contractors to put Bedrock Robotics' fully autonomous excavators to work on real projects.

Rachael and I met at ENR FutureTech and have been trying to get this conversation on the calendar ever since. Turns out she was a little busy.

So I asked the first thing any field guy would ask: what does the crew think?

Her answer was honest. The early feedback from the field, which she says they expected, was "that's not as fast as a human." She's fine with that. Speed wasn't the line they just crossed. Running safely with an empty cab was. Speed comes next.

In this episode we get into:

• How Zachry is measuring the machine: one autonomous excavator working next to human crews on the same job, moving the same dirt
• Why they bid the job with people only and added the machine after
• The real labor problem, which isn't empty seats but the decades of skill that are hard to replace
• Remote sites, 12-hour shifts, months away from family, and where a night shift with no operator could fit
• Why schedule and safety outrank price on the big data center builds, and how strange that feels coming from low-bid highway work
• What separates the tech companies that make it in construction from the ones that never owned a pair of boots
• What she wants next: a whole crew of machines you hand a goal to, not a list of instructions.

I really respect the work: it took hundreds of calls, texts and meetings to get one machine on site. It also took a field crew already working six days a week. The technology is the headline. The people who made room for it are the story.

Real People. Real Work. Real Talk.

Owner hires an architect. Architect draws something beautiful. GC bids it out with two weeks and zero contact with the person who drew it. The trades get three days and a set of drawings with the RFIs still baked in. Then everybody acts surprised when the change orders start.

Viyas Sundaram, CEO of STACK Construction Technologies, calls it what it is on this episode. His words: "Stuff rolls downhill in that configuration." (He cleaned it up for us. Then he didn't.) Viyas started as free labor on his immigrant parents' dream home, ran GoCanvas through its sale to Bluebeam's parent company, and took over STACK in March. His argument is that the funnel doesn't just dump risk on the trades, it throws away the most valuable knowledge on the project: how the thing actually got built. Robert's scrubber story, where an architect walked him into a room with concrete on all six sides and pointed, is the whole problem in one anecdote.

Also in this one:

• Why Viyas says the trades are "in the business of getting shit done," not bidding, and why nobody became a carpenter for their Excel skills
• The speed-versus-accuracy fight in AI takeoffs: big contractors carry buffer to absorb an 80% answer; the three-person shop on thin margins can't
• "Construction is the art of layering on top of imperfections," and why the trades run out of layers first
• Why an "AI" sticker on the box impresses nobody in the field
• Data ownership: STACK's stance that your own project history should train your estimates, not public data, unless you opt in
• Robert's pushback on "value engineering" (there's no such thing; call it cost engineering)
• Viyas's take on where it all ends: when the job site feeds back to the drawings, "there is no pre when it's a loop"

If you've ever been handed a bid package with a question you already knew the answer to, this one's for you.

STACK: stackct.com. Viyas Sundaram is on LinkedIn.